Working Paper Series: Special Edition of 2016 to 2018 Interns

II. Time Intervals Specification The Im-Pesaran-Shin panel unit root test ( Im et al., 2003 ) was applied to check the annual series for stationarity between 2000 and 2015. . Results revealed that the null hypothesis for non- stationarity is rejected for all variables except log per capita GDP ( See Appendix, Table 3 ). This non-stationarity of variables supported the choice to use non-overlapping time intervals rather than the annual data set. The panel data was therefore transformed using four 4-year average periods (2000-2003, 2004-2007, 2008-2011 and 2012-2015) 34 .a. The use of four-year averages was helpful because yearly averages reduce the influence of short term economic shocks, and it avoids the problem of non-stationarity in variables which may produce biased regression results. Furthermore, compared to the sixteen-year annual data, the use of four-year intervals allowed GMM estimations to be more robust given the time dimension necessary to acquire more precise results. . Summary Statistics Figure 1 shows that the higher the per capita GDP the lower the remittance inflow. It also shows a significant clustering of countries in the top left hand corner which are mainly using a fixed exchange regime, except for El Salvador and Guyana. III.

Figure 1. Per Capita GDP (natural log) scattered against Remittances.

34 Non-overlapping time intervals were also used in empirical studies by Bonnefond (2014), Narayan et al (2011), Zghidi et al (2016), Brun et al (2002) and Giuliano and Ruiz-Arranz (2005).

128